Tian Dong (Shanghai Jiao Tong University), Minhui Xue (CSIRO's Data61), Guoxing Chen (Shanghai Jiao Tong University), Rayne Holland (CSIRO's Data61), Yan Meng (Shanghai Jiao Tong University), Shaofeng Li (Southeast University), Zhen Liu (Shanghai Jiao Tong University), Haojin Zhu (Shanghai Jiao Tong University)

Open-source Large Language Models (LLMs) have recently gained popularity because of their comparable performance to proprietary LLMs. To efficiently fulfill domain-specialized tasks, open-source LLMs can be refined, without expensive accelerators, using low-rank adapters. However, it is still unknown whether low-rank adapters can be exploited to control LLMs. To address this gap, we demonstrate that an infected adapter can induce, on specific triggers, an LLM to output content defined by an adversary and to even maliciously use tools. To train a Trojan adapter, we propose two novel attacks, POLISHED and FUSION, that improve over prior approaches. POLISHED uses a superior LLM to align naïvely poisoned data based on our insight that it can better inject poisoning knowledge during training. In contrast, FUSION leverages a novel over-poisoning procedure to transform a benign adapter into a malicious one by magnifying the attention between trigger and target in model weights. In our experiments, we first conduct two case studies to demonstrate that a compromised LLM agent can use malware to control the system (e.g., a LLM-driven robot) or to launch a spear-phishing attack. Then, in terms of targeted misinformation, we show that our attacks provide higher attack effectiveness than the existing baseline and, for the purpose of attracting downloads, preserve or improve the adapter’s utility. Finally, we designed and evaluated three potential defenses. However, none proved entirely effective in safeguarding against our attacks, highlighting the need for more robust defenses supporting a secure LLM supply chain.

View More Papers

VeriBin: Adaptive Verification of Patches at the Binary Level

Hongwei Wu (Purdue University), Jianliang Wu (Simon Fraser University), Ruoyu Wu (Purdue University), Ayushi Sharma (Purdue University), Aravind Machiry (Purdue University), Antonio Bianchi (Purdue University)

Read More

Heimdall: Towards Risk-Aware Network Management Outsourcing

Yuejie Wang (Peking University), Qiutong Men (New York University), Yongting Chen (New York University Shanghai), Jiajin Liu (New York University Shanghai), Gengyu Chen (Carnegie Mellon University), Ying Zhang (Meta), Guyue Liu (Peking University), Vyas Sekar (Carnegie Mellon University)

Read More

DRAGON: Predicting Decompiled Variable Data Types with Learned Confidence...

Caleb Stewart, Rhonda Gaede, Jeffrey Kulick (University of Alabama in Huntsville)

Read More

“Where Are We On Cyber?” – A Qualitative Study...

Jens Christian Opdenbusch (Ruhr University Bochum), Jonas Hielscher (Ruhr University Bochum), M. Angela Sasse (Ruhr University Bochum, University College London)

Read More